Georgios Tsoukas

Modifying train.py as follows worked for me: ``` # Training loop. For each batch... best_perf = -1.0 # larger is better for batch in batches: x_batch, y_batch = zip(*batch) train_step(x_batch,...

> I think that would be interesting! I think the downside is > > 1. Would make it more complex Fully agree, probably the key reason not to do it....

I did some benchmarks with a 150k vector dataset of 64 dimensions with inner product.(will have to repeat on a public dataset in order to be able to share details)....

Thank you @yurymalkov , it would be great to have this functionality. Here are some benchmarks demonstrating on a public dataset that the filtering functionality for medium to large fractions...

Thank you so much for the review @dyashuni! I don't know what to do about the last comment (line 613). Happy to dig deeper if you could give me a...

@dyashuni, thanks a million times for the patch! I have applied the patch as-is. All my functional and performance tests look good after application of the patch. For someone not...

> In python when we add items (`add_items` function) we can also specify corresponding labels/ids via `ids` parameter. If the `ids` parameter is not specified then items get sequential labels/ids...